METHOD / NEXT
Judge-dependent safety gains and model-specific helpfulness costs of evidence-sufficiency prompting in clinical LLMs
Across 1,200 paired cells, evidence-sufficiency prompting reduced judge-rated overconfidence, but effect size depended on the judge and helpfulness costs were model-specific. The author explicitly states that this does not establish deployment readiness.
Source: arXiv:2607.18086arXiv:2607.18086 | arXiv | 2026Judge-dependent safety gains and model-specific helpfulness costs of evidence-sufficiency prompting in clinical LLMsKoyar AfrasyabAcross 1,200 paired cells, evidence-sufficiency prompting reduced judge-rated overconfidence, but effect size depended on the judge and helpfulness costs were model-specific. The author explicitly states that this does not establish deployment readiness.Open paper
METHOD / NEXT
GigaPath-Flash and GigaTIME-Flash: Efficient Pathology Foundation Models for Whole-Slide and Tumor Microenvironment Analysis
The authors introduce compact whole-slide pathology and tumour-microenvironment models and report near-teacher performance at lower compute. Open weights improve reproducibility, while independent external validation remains necessary.
Source: arXiv:2607.18218arXiv:2607.18218 | arXiv | 2026GigaPath-Flash and GigaTIME-Flash: Efficient Pathology Foundation Models for Whole-Slide and Tumor Microenvironment AnalysisNaoto Usuyama et al.The authors introduce compact whole-slide pathology and tumour-microenvironment models and report near-teacher performance at lower compute. Open weights improve reproducibility, while independent external validation remains necessary.Open paper